Cooperative Transportation Without Prior Object Knowledge via Adaptive Self-Allocation and Coordination

1. Problem Statement

The paper addresses the challenge of multi-agent cooperative transportation in environments where the properties of the target objects (cargos) are unknown a priori.

• Limitations of Existing Methods: Traditional approaches often require precise prior knowledge of object location, size, shape, and weight. They frequently rely on fixed team sizes or centralized coordination, which limits robustness in dynamic, real-world scenarios where objects may be newly discovered, irregular, or partially observed.
• Core Challenge: How can a team of agents autonomously detect unknown objects, determine the necessary number of agents required for transport based on object size, and form a stable, balanced enclosure without centralized assignment or global information?
• Constraints: Agents operate with limited sensing ranges, communicate only with local neighbors, and must avoid collisions while maintaining safe inter-agent distances.

2. Methodology

The authors propose a density-driven, decentralized framework that integrates three key components: Adaptive Density Functions, Centroidal Voronoi Tessellation (CVT), and Control Barrier Functions (CBF).

A. Adaptive Object Detection and Density Generation

• Local Sensing: Agents rely solely on local sensors to detect objects.
• Dynamic Density Function ($\phi(q, t)$): Upon detecting an object, an agent generates a Gaussian attraction field centered at its position. The overall density function for the workspace is a superposition of these individual fields:
  $$\phi(q, t) = \phi_0 + \sum_{i=1}^{N} \rho_i(t) \cdot \text{Gaussian}(q; \mu_i, \sigma_i)$$
  • $\rho_i(t)$ is a binary switch (1 if agent $i$ detects an object, 0 otherwise).
  • The density peaks correspond to the locations of agents detecting the object.
• Self-Allocation Mechanism: The density function naturally scales with the number of detecting agents. Larger objects are detected by more agents (or agents detect them over a wider area), creating a higher cumulative density. This implicitly signals that more agents are needed for transport, enabling adaptive team formation without explicit task assignment.

B. Centroidal Voronoi Tessellation (CVT) for Spatial Organization

• Goal: To distribute agents evenly around the detected object to form a stable "cage" or enclosure.
• Mechanism: The system minimizes a locational cost function $H(p, t)$ defined over the Voronoi cells of the agents.
  $$H(p, t) = \sum_{i=1}^{N} \int_{V_i} \|q - p_i\|^2 \phi(q, t) dq$$
• Control Law: Agents move toward the centroid ($c_i$) of their respective Voronoi cell, calculated with respect to the density function $\phi$.
  $$u_{nom} = -k(p_i - c_i)$$
• Result: This drives agents to a locally optimal configuration where they are uniformly distributed around the high-density regions (the objects), ensuring balanced force application for transport.

C. Control Barrier Functions (CBF) for Safety and Stability

• Problem: Pure CVT control might lead to agents clustering too closely or unevenly on one side of a large object, causing instability or collisions.
• Solution: A CBF-based constraint is added to the control input via a Quadratic Program (QP).
  • Safety Set: Enforces a minimum distance $d_{min}$ between any pair of agents ($i, j$).
  • Constraint: $\dot{h}_{ij}(t) + \gamma h_{ij}(t) \geq 0$, where $h_{ij} = \|p_i - p_j\|^2 - d_{min}^2$.
• Outcome: The CBF modifies the nominal CVT control input to guarantee forward invariance of the safety set. This prevents agent clustering, ensures a symmetric distribution around the cargo, and maintains collision-free motion during close-proximity cooperation.

3. Key Contributions

1. Zero-Prior-Knowledge Framework: The system operates without any pre-existing map, object model, or knowledge of object size/location. It relies entirely on local sensing and reactive behaviors.
2. Adaptive Self-Allocation: The framework automatically scales the number of participating agents to the size of the cargo. Larger objects generate stronger/more complex density fields, naturally attracting more agents to the task.
3. Integration of CVT and CBF: The novel combination of CVT (for optimal spatial distribution) and CBF (for safety constraints) solves the dual problem of forming a balanced enclosure while preventing dangerous clustering.
4. Decentralized Coordination: The approach requires no central coordinator; all decisions are made based on local interactions and the shared density field.

4. Simulation Results

The authors validated the framework through numerical simulations in a 2D bounded workspace with multiple heterogeneous cargos (different sizes).

• Scenario: Agents started in a compact formation with no knowledge of cargo locations.
• Process:
  1. Agents dispersed to explore.
  2. Upon detection, density fields activated, and agents began converging.
  3. Agents self-organized into balanced enclosures.
• Observations:
  • Size-Dependent Allocation: The larger cargo successfully attracted a larger team (7 agents) compared to the smaller cargo (5 agents).
  • Stable Transport: Agents formed symmetric formations around the objects and transported them toward boundaries without collisions.
  • Robustness: The system handled the transition from exploration to transport smoothly, with no excessive oscillations or agent aggregation.
  • Voronoi Adaptation: The Voronoi partitions dynamically adjusted as agents moved, maintaining balanced spacing throughout the process.

5. Significance and Future Work

• Significance: This work bridges a critical gap in multi-robot systems by enabling cooperative transport in highly uncertain environments. It moves away from rigid, pre-programmed strategies toward flexible, bio-inspired self-organization. The ability to handle irregular shapes and unknown sizes makes it highly applicable to real-world logistics, disaster response, and construction.
• Future Work: The authors plan to incorporate more realistic physical properties (e.g., object weight and friction) and validate the framework using physical multi-robot hardware experiments.

In summary, the paper presents a robust, scalable, and fully decentralized solution for cooperative transportation that leverages density-driven attraction, geometric optimization (CVT), and safety constraints (CBF) to achieve adaptive, collision-free transport of unknown objects.